Type I error rate with different randomisation and analysis methods. (A) shows the type I error rate for different values of the interaction term between the two balancing factors. (B) shows the type I error rate for different values of P(X12 = 1) (which indicates the proportion of patients in the interaction group i.e. with both prognostic factors present). Two methods of randomisation were used (Min = minimisation, SPB = stratified permuted blocks), and two methods of analysis were used (covariate adjusted indicates the two balancing factors were used as covariates in a regression model; stratified analysis denotes that the strata formed from the combinations of the two balancing factors were entered as covariates in a regression model).